MOTA baseline -> method
87.1 -> 90.4
Computer Vision · UAV Swarm Detection and Tracking · Formation Constraints
V-USDT
V-USDT: Vision-Based UAV Swarm Detection and Tracking by Leveraging Swarm Formation Constraints
Overview
V-USDT addresses vision-based UAV swarm detection and tracking using formation-aware reasoning and temporal consistency.
Problem Statement
UAV swarms create detection and association challenges due to small object scale, similar appearance, clutter, and frequent interaction between targets.
Motivation
Tracking quality depends not only on frame-level detections but also on maintaining identity and trajectory consistency over time.
Methodology
V-USDT integrates vision-based detection with temporal association and swarm formation constraints to reduce identity ambiguity across time.
Key Contributions
- Vision-based UAV swarm detection and tracking.
- Formation-aware reasoning for multi-target perception.
- Temporal consistency for robust swarm-level tracking.
Experimental Setup
Input modality: vision-based UAV swarm video / image observations. Task: detection and tracking of UAV swarms with formation-aware temporal association.
Quantitative Results
IDF1 baseline -> method
91.1 -> 93.1
Misses baseline -> method
78.24 -> 63.29
Demo / Media
My Role
My contributions include task framing, method integration, experimental design, analysis, and manuscript writing.
Related Publication
M. H. Rahman and S. Madria. "V-USDT: Vision-Based UAV Swarm Detection and Tracking by Leveraging Swarm Formation Constraints." IEEE MDM, 2025.